# A First Look at System One Decision Models for Malware Detection

DevFeed: [A First Look at System One Decision Models for Malware Detection](<https://devfeed.tech/articles/a-first-look-at-system-one-decision-models-for-malware-detection-137605.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.20699>)

Author: Pier Mattia Basciano, Giovanni Farina, Salvatore Monteleone

Published: 2026-11-17T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>), [arXiv: Cryptography and Security](<https://devfeed.tech/sources/arxiv-cryptography-and-security.md>), [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [Malware](<https://devfeed.tech/topics/malware.md>), [Detection engineering](<https://devfeed.tech/topics/detection-engineering.md>), [virtual machines](<https://devfeed.tech/topics/virtual-machines.md>), [Windows](<https://devfeed.tech/topics/windows.md>), [Large Language Model](<https://devfeed.tech/topics/llm.md>), [Feature Engineering](<https://devfeed.tech/topics/feature-engineering.md>)

Tags: [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cr](<https://devfeed.tech/tags/cs-cr.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [decision-models](<https://devfeed.tech/tags/decision-models.md>), [detection](<https://devfeed.tech/tags/detection.md>), [large-language-models-llms](<https://devfeed.tech/tags/large-language-models-llms.md>), [malware](<https://devfeed.tech/tags/malware.md>), [malware-detection](<https://devfeed.tech/tags/malware-detection.md>), [system-one](<https://devfeed.tech/tags/system-one.md>)

## AI overview

The study evaluates System One decision models for Windows PE malware detection and proposes S1Mal, which combines answers to 32 behavioral questions with CLM-8B's embedding-based score. Binaries are processed with Ghidra and capa into a canonical text state; models are scored using a single verdict probability (A0), weighted atomic answers (A1), or embedding-based heads (A2). On BODMAS, A0 AUROC ranges from 0.4025 to 0.8590 across 13 models, while A1 improves AUROC by an average of 54.61%; A2 reaches 0.9943 for CLM-8B. S1Mal reaches 0.9946 AUROC and detects 59.15% of malware at 0.1% FPR on the test ROC curve, compared with 46.59% for A2. The abstract also reports evaluation on SOREL-20M and matched benign files, but gives no results for those settings.

## Source excerpt

arXiv:2611.20699v1 Announce Type: cross Abstract: Malware detection needs scores calibrated at a fixed false positive rate (FPR) and decisions an analyst can read: feature-based detectors give the first, large language models (LLMs) on decompiled code the second, through generated text of uncalibrated confidence. System One decision models answer a typed question about a text with a probability over its declared answers in one forward pass. We present the first evaluation of these models for Windows Portable Executable malware detection, analyze how they decide, and propose S1Mal, an ensemble on the typed answers and the embedding of CLM-8B. In an isolated virtual machine, Ghidra and capa decompose each binary, and the host renders the records into a canonical textual state, a digest of headers, imports, MITRE ATT&CK capabilities and strings, optionally with decompiled functions, without class shortcuts. Three readouts score each model on the state: A0, the probability of one verdict question; A1, 32 atomic behavioral questions weighed by gradient-boosted trees; A2, heads on the model's state embedding. S1Mal feeds gradient-boosted trees with the atomic answers and the A2 score of CLM-8B. We evaluate CLM-8B, ten open decision models and two LLMs read through logits on BODMAS temporal splits, SOREL-20M out-of-distribution malware and 14,534 matched benign files. On BODMAS, A0 wastes the static evidence: its area under the receiver operating characteristic curve (AUROC) spans 0.4025-0.8590 across 13 models. A1 recovers it, raising every model's AUROC by +54.61% on average; A2 lifts CLM-8B by +0.2749 to 0.9943, above a combiner on digest facts (0.9901), and reaches 0.9853-0.9948 on nine open models. S1Mal reaches 0.9946 and detects 93.47% of the malware at a realized FPR of 1.45%, and 59.15% at 0.1% FPR on the test ROC curve, against 46.59% with A2. S1Mal gives analysts calibrated decisions, each with the atomic answers behind it.